WHY SMALL CHILDREN CANNOT CHANGE
LANGUAGE ON THEIR OWN: SUGGESTIONS
FROM THE ENGLISH PAST TENSE
JOAN L. BYBEE
DAN I. SLOBIN
The striking parallel between children's morpho-phonemic or analogical innovations and the morpho-phonemic changes that can be
observed in the history of languages has long been noted. (Hooper
19 79 and Andersen 1980 are recent examples). Some linguists have
proposed that the explanation for this parallel is that children are
responsible for instigating this type of change in language. However,
for children's innovations to have a permanent effect on the language, they must be maintained into adulthood. Andersen 1973
hypothesizes that as speakers reach adulthood, they maintain basically the same system established during the language acquisition
period, but adapt it to the norm in various ways (e.g. by memorizing exceptions). To the extent that a succeeding generation arrives
at a similar internalized grammar, their innovations will seem acceptable to the adult, and the adult norm will gradually change. Under this hypothesis, the reason that children's speech seems to reflect morpho-phonemic change so well is that children are freer to
be more creative and to produce more innovations. Thus, adult innovations might bear the same resemblance to historical change as
children's innovations do, but adult innovations are much rarer,
and certainly less well studied.
This hypothesis can be tested by examining carefully the innovations of children and adults and comparing them with on-going
morpho-phonemic changes. We undertook such a comparison using
the English past tense, concentrating on the treatment of the irregular verbs. We compared data from three age groups: (1) From 31
pre-school children, aged one and a half to five years, we have spontaneous speech data, and from 20 three-and four-year old children
we have the results of an elicitation task; (2) From 15 third-grade
children between the ages of eight and a half and ten years we have
elicited past tense forms from a sentence completion task; (3) We
also elicited innovations in past tense formation from 40 universi-
30
SMALL CHILDREN AND LANGUAGE
JOAN L. BYBEE & DAN I. SLOBIN
ty-aged adults. In order to elicit innovations from adults, we had to
create conditions of pressure and fatigue. We did this by asking the
subjects to listen to a list of 180 verbs, and to supply a past tense for
each base form they heard. They were asked to respond as quickly
as possible, and the experimenters were urged to put as much pressure for speed on the subjects as possible. This experiment and the
other data sources are described in detail in Bybee & Slob in 1981.
Our question is, then, if we compare morpho-phonemic innovations of pre-school children, third-grade children and adults, which
of these parallel most closely documented on-going changes in the
system? The comparison revealed three areas in which the innovations of all three groups paralleled historical change, and one area in
which the pre-school children's innovations reflected universal
tendencies while the innovati~ns of the two older groups reflected
English-specific tendencies. We begin by describing the three areas
in which all three groups parallel on-going changes.
The first point concerns the correlation between the percentage
of time a subject regularized a verb (e.g. weep, weeped) and the frequency of that verb in discourse. It is well known that in the history
of languages, the lexical items that are the most prone to analogical
leveling or regularization are the least frequent items. In Hooper
1976a it was argued that this fact (established statistically there for
the English irregular verbs), points to imperfect learning or child
language as the source of this sort of change, since irregular items
that are rote learned can only be stable if they are frequent enough
to be learned well. This argument, however, may also be extended
to adults, since it is possible that in order for adults to maintain an
irregular item without regularizing it, the item must continue to be
used at a certain level of frequency. Thus we might expect to find
all three age groups regularizing the least frequent verbs to a greater
extent. This is precisely what our data show. The trend is strongest
for the pre-school children. A significant negative rank order correlation ( -.6 7) was found between the number of times the caretakers or adults used the verbs in the spontaneous speech samples and
the number of times the verbs were regularized by the children. The
regularization made by the older two groups of subjects were correlated with the frequencies of past tense forms according to Kutera
and Francis 196 7. Negative correlations that were significant or
that approach significance were found for four out of the six verb
classes that were large enough to test (Bybee & Slob in 1981 ). These
31
data show that the same relationship between frequency and regularization exists in speakers of all ages as well as in historical change.
Thus it is not just young children's imperfect learning that prompts
historical regularization, but rather the general weakness of a low
frequency irregular item for speakers of all ages.
The second point of agreement among all three age groups and
history concerns the class of verbs which includes send, sent; bend,
bent; lend, lent; spend, spent; and build, built. These verbs form
their past tense by changing a final d tot. All three age groups regularized these verbs more often than the average. This was especially
striking for the third-graders, who regularized this class of verbs 53%
of the time, while their average of regularizations for verbs of all
types was only 1 7%. The preschoolers regularized this class 57% of
the time, compared to an overall average of 39%, and the adults
showed a much smaller margin, regularizing this class 20% of the
time compared to an overall1 7%.
The difficulty with this class of verbs seems to be that it employs
a process quite distinct from any other used to form the past tense
in English. These verbs neither undergo a vowel change nor add a
suffix, but rather devoice a final consonant, which makes them unique in all of English morphology. Furthermore our subjects's difficulty with these verbs reflects an historical trend toward eliminating this class altogether. The following verbs are listed by Jespersen
1942 as belonging to this class:
( 1)
bend
lend
send
spend
build
blend
geld
gird
gild
wend
rend
shend 'to shame'
The five verbs in the first column are the viable members of this
class. The next five have regularized in American English. Note further that of these five, only blend is in common use, the others, including rend from the last column are dangerously near passing out
of the language, and shend already has in American English.
Thus this class of verbs is losing members to the extent that it probably should no longer be considered a class, and our subjects' regularizations of these verbs reflects this movement. Note further that
while the usual way for an irregular class to lose members is by regularizations, this class is losing members by lexical attrition. That is,
some of the verbs have fallen into disuse while new verbs are substi-
32
JOAN L. BYBEE & DAN I. SLOBIN
SMALL CHILDREN AND LANGUAGE
tuted for them. This trend was also observable among our adult
subjects, four of whom gave loaned as the past tense of lend.
The third area of convergence to be discussed concerns the verb
classes following the patterns of sing, sang, sung and sting, stung.
Here we are not concerned with regularization errors but rather
with errors of incorrect vowel change. Consider now these two related classes of verbs, grouped according to the final consonant:
groups with historical innovations.
The third-graders only made 38 innovations of incorrect vowel
change, and 32 of them involved these two types of verbs. The
adults made 85 such innovations and nearly one half, 41 of them,
involved these classes of verbs. It is clear, then, that these are the
strong verb classes ·where the action is in the contemporary language.
The most common type of innovative form was one in which a
verb of the sing, sang class was treated as though it belonged to the
sting, stung class. In other words a past tense in /A/ was given. The
number of times this occurred is registered in ( 4 ).
(2)
m
n
swim
swam
begin
began
run
ran
ng
ring
rang
sing
sang
spring
sprang
nk
drink
drank
shrink
shrank
sink
sank
stink
stank
*not originally an OE strong verb.
(3)
swum
begun
run
rung*
sung
sprung
drunk
shrunk
sunk
stunk
n
spin
spun
win
won
ng
cling
clung
fling
flung*
sling
slung*
sting
stung*
string
strung*
swing
swung
wring
wrung
hang
hung*
nk
slink
slunk
k
strike
struck*
stick
stuck*
sneak
snuck*
g
dig
dug*
drag
drug*
*not originally an OE strong verb.
Innovations of incorrect vowel change were not recorded for the
pre-school children, so our discussion here involves only the thirdgrade children and the adults. Our point here will be to demonstrate the convergence of the innovations made by these two
33
3rdgraders
adults
9
shrunk
5
shrink
6
rung
5
ring
1
swim
swum
5
2
begun
begin
2
6
drunk
2
drink
*
7
sunk
sink
*Not on 3rd graders test.
(For the adults, these are out of a total of 20 responses; for the 3rd graders, out
of 15 responses.)
(4)
These responses parallel the historical trend which is to eliminate
the distinction between the past tense form and the past participle
form, in favor of the past participle form. This is precisely how the
sting, stung class was formed originally: all the strong verbs in this
class originally had three principle parts, e.g. spring, sprang, sprung;
spin, span, spun. Thus the non-standard forms that adults use under
pressure are the same as the non-standard forms the third-graders
use, and the same forms that we would predict would come into use
if past trends continue. Once again, the adult innovations converge
with historical trends, much as children's forms do.
A second type of innovation involving the sting, stung class that
parallels historical trends is the attraction of new members to this
class. All the forms asterisked in (3), (10 out of 16) have been
brought into this class since the Old English period. This is clearly
the most productive of strong verb classes. We are not surprised
that our subjects also added new members to this class by producing the forms in ( 5 ).
34
JOAN L. BYBEE & DAN I. SLOBIN
(5)
bring
think
clink
streak
brung
thunk
clunk
struck
eat
ut
adults
1
0
1
4
SMALL CHILDREN AND LANGUAGE
3rdgraders
2
1
*
*
0
*Not on 3rd graders' test
Again, the trend evidenced in these responses parallels neatly the
long-standing historical trend.
Our subjects also produced some counter-historical innovations.
In these they treated verbs of the sting, stung class as though they
belonged to the sing, sang class, and gave them past tenses in /tr/.
They also treated one weak verb clink as though it were a member
of the n"ng, rang class. Consider the examples in ( 6):
(6}
adults
string
sting
slink
clink
strang
stang
slank
clank
clanked
*Not on 3rd graders' test
2
2
1
1
3
3rdgraders
1
*
*
Parallel historical innovations have not occurred, except in the case
of ring, rang, rung which was originally weak 0 espersen 1942). In
accord with the historical trend, such innovations are much rarer in
our data than the innovations that lead to new members of the sting
stung class.
The three points just mentioned, then, are the major trends identifiable across all three age groups, and these are precisely the
trends that are parelleled in the ongoing changes in American English. It seems reasonable to conclude, then, that young children are
not the only, and perhaps not the primary instigators and perpetrators of morpho-phonemic change. Stronger evidence for this point
come from innovation in verbs with a zero past tense suffix, e.g. cut,
h£t, set, where the older children and adults seem to reflect ongoing changes in the system better than the younger, pre-school
children do. We tum now to a discussion of these verbs.
When we compared the percentage of regularizations on verbs
that take a zero-suffix, i.e. the hit class, across age groups, we find a
striking anomaly: while the percentage of regularizations overall
35
drops dramatically between pre-school children and third-graders,
the percentage of regularizations on the h£t-class remains virtually
the same across age groups. The result is that for pre-schoolers the
regularizations in this class are below their overall average, i.e. 29%
for the h£t-class versus 39% overall, but for the third-graders and
adults, the regularizations in the class are above average, 27% compared to 1 7% overall for both groups. The question is, then, why do
subjects get worse at dealing with this type of verb rather than better? The answer we suggest in Bybee & Slobin 1981 is the following.
The strategy that the youngest children start with in trying to analyse past tense formation in English, does not involve necessarily
the postulation of a suffixation process, but may involve simply the
observation that past tense verbs end in an alveolar stop. (Such a
generalization is called a schema.) Thus they know what the end
product must be like, but they have not formulated the suffixation
process as a necessary means for arriving at the end product. (They
have formulated a product-oriented modification, in the terms of
Zager 1980). Verbs such as h£t, put and cut already contain the
phonetic material necessary for signalling past tense, that is, the final t. These youngest children are content that past tense is adequately signalled by this t and so are less likely to add a suffix to these
verbs than to some others.
The pre-schoolers' strategy reflects a universal tendency, called
'affix-checking' by Menn & MacWhinney (forthcoming) and 'morphological haplology' by Sternberger 1980, to avoid adding affixes
to words or stems that appear already to contain these affixes. Indeed, it is no accident that the only verbs in English chat are unchanged in the past are verbs that end in t and d. Examples of similar phenomena in other languages are documented by Menn & MacWhinney (forthcoming) and Sternberger 1980. There is further evidence that it is this strategy that accounts for the pre-schoolers'
relative success with these verbs. When Berko 195 7 tested children
with nonce forms, she found that children preferred to add no past
tense suffix to verbs ending in tor d, and similarly, they preferred
to add no plural suffix to nouns ending in a sibilant. The strategy,
then, is not limited to verbs. In the pre-school elicitation data wereport on in Bybee & Slob in 1981, there is a highly significant tendency not to add a suffix even to regular verbs, such as melt, that
end in t or d. (For further evidence of this strategy, see Kuczaj
1978).
36
JOAN L. BYBEE & DAN I. SLOBIN
SMALL CHILDREN AND LANGUAGE
The third-graders and adults differ from the pre-schoolers in that
their major strategy is the suffixation process: rather than operating
with the principle that a past tense verb ends in t or d, they operate
with the principle that a past tense form is a base form with t of d
added to it. Thus their tendency is to make many errors of adding
ed to verbs that should not change in the past tense. This does not
mean that the child's strategy is not still operative in adults-it is to
a limited extent. Where the third-graders and adults produced past
tense forms without changing the base (where a change should have
been made), this was more than twice as likely to be with a verb ending in t or d than with a verb ending in some other segment. But
overall the strength of the suffixation process is evident in the relatively high percentage of regularization of verbs of the h£t-class.
Now compare the history of this class of verbs. They arose quite
naturally as the weak preterite suffix de assimilated to final t to give
OE preterites such as sette for settan. Subsequent final vowel
deletion produced identical present set and preterite set. A class of
verbs with primarily lax vowels and final t and d that undergo no
change in the preterite was born. This class has attracted new members from the strong verbs and also from borrowings: e.g. b£d, bust,
sl£t, cost Uespersen 1942:28). The class enjoyed a period of productivity that culminated, according to Jespersen, early in the Modem English period. At that point some verbs that were previously
unchanging began to regularize, e.g. fast, start, l£ft, fret. The trend
seems to be continuing with e.g. roast and sweat regularizing as
well. It appears, then, that though this class has enjoyed some productivity, due perhaps to the universal tendency mentioned above,
the regular suffixation process has continued to gain strength and
overrides the productivity of this class in current English.
If this sketch of the history of the h£t-class correctly interprets
current trends, then the third-grade children's and the adults' innovations are more in accord with on-going changes in the system
that the pre-school children's formations are. While the pre-school
children do regularize all verbs, including the hit-class, their tenden·
cy to regularize these less compared with the older groups' tendency to regularize them more, suggests that the pre-school children's
innovations accord more with universal tendencies while the older
groups manifest more English-specific tendencies.
The conclusion that must be drawn from the facts is that there is
nothing particularly special about the relation between small child-
ren 's innovative forms and morpho-phonemic change. The innovation of older children and adults, though perhaps rarer, where they
can be elicited, may also serve as predictors of change. In fact in
some cases where adult innovations differ from early child innovations, such as with the hit-class, the adults and older children, who
are in better command of the entire system, innovate in ways that
manifest more precisely the on-going changes in the system. Thus
it appears that both socially and linguistically the older children
and adults are in control of morpho-phonemic changes.
This is not to say that the study of early child language is of no interest to historical linguists. Quite the contrary, since it is only in
early child language that universal principles or morpho-phonemic
organisation are manifested most clearly. Here children not yet under the influence of an internalized language-specific system reveal
certain very general operating principles for approaching morphophonemics (Slobin 1973, MacWhinney 1978). These universal principles will underly the adult system. The child's task is to sort out
how conflicts among principles, and conflicts between principles
and input data, are to be resolved. The interplay between conflicting principles places the seeds of change in these universal principles. In the example we have just discussed, the h£t- class of verbs
developed fortuitously by regular sound change, but was stabilized
and made productive, because these verbs fit in with the universal
principle which allows zero-affixation in case a base already contains the phonetic material of the affix. This principle was previously manifested nowhere else in English. Its source at this point must
have been the operating principles of early child language. On the
other hand, it conflicts with the operating principle that clear segmentation of markers should be maximized. The current resolution
of this conflict is in favor of the latter principle which is realized as
a suffixation rule. But the realization that both principles are
operative in early child language helps us to understand the changes
that have taken place over the centuries in English. Ourdatasuggest,
then, that current changes in a language will be better reflected in
adult innovations, and that adults are actually responsible for carrying out morpho-phonemic change. Young children, on the other
hand, give a better indication of the full range of possible changes
and are an important source of information for a theory of morphophonemic change.
37